You are the only one talking about you

The most common content plan we are shown goes like this: publish two articles a week on the company blog, target the questions customers ask, add FAQ schema, wait. Six months later the articles are good, the schema validates, and AI engines are still recommending a competitor who publishes half as much.

The plan is not wrong. It is incomplete in a way that classic SEO rarely punished and AI search punishes hard. It assumes the engine will take your word for it.

What the citation studies keep finding

Between 2024 and 2026 several groups analysed hundreds of millions of AI citations across ChatGPT, Perplexity, Gemini, AI Overviews and AI Mode. The details differ; the shape does not. A small set of domains dominates. Reddit, Wikipedia, YouTube and LinkedIn appear in every top ten. Established editorial brands and review platforms fill most of the rest. One consolidated index put the top 15 domains at around 68 percent of all citations, with nothing like that concentration in Google's organic results for the same queries.

Brand-owned pages appear, but at a fraction of the rate their owners assume. Within ChatGPT the picture shifted again after late 2025 towards Wikipedia, LinkedIn and editorial sources, and LinkedIn in particular rose sharply on the back of published posts and articles rather than profile pages.

The mechanism is not mysterious. An engine assembling an answer about which business to trust is looking for corroboration. Your own page saying you are excellent is a claim. A review platform, a local paper, a trade body, a LinkedIn article by a named person and a Reddit thread all describing you consistently is evidence. Engines are built to prefer evidence.

The owned-only trap

There is a specific, measurable version of this. One 2026 analysis found brands relying on owned content alone achieved around 18 percent AI coverage for their target questions; adding earned coverage took the same brands to roughly 58 percent. Treat the exact numbers loosely; the gap is the point. Publishing more on your own site has sharply diminishing returns while nobody else mentions you.

It gets worse when the owned content is hard to read. When an engine cannot parse your pricing page, it does not give up on the question; it answers it from G2, Trustpilot, Yell or a competitor. Owned-only businesses are not just under-cited; they are described by third parties they did not choose.

What "earned" means at your scale

This is where most advice loses small businesses, because it is written for brands with PR budgets. Earned coverage for a heating engineer in Hailsham or a 12-person accountancy in Eastbourne looks different, and it is more achievable than it sounds:

• Reviews with detail on Google and on the one platform your sector trusts (Checkatrade, Trustpilot, Yelp, G2). Yelp's data now feeds ChatGPT's local answers under a 2026 agreement.
• LinkedIn publishing by named people, not just a company page. An article a month from the founder answering a real customer question is the single highest-leverage move for ChatGPT visibility in the 2026 data.
• YouTube explainers: five to ten short videos answering the questions you get on the phone, with transcripts. YouTube became the most-cited social source across engines in early 2026.
• Honest Reddit and forum participation where your customers ask questions, with your name and role visible and no link-dropping.
• Local and trade press, the Chamber of Commerce, sector bodies, supplier partner pages. Boring, and exactly the sources engines already cite for local and B2B questions.
• One piece of original data a year that others will quote, with your name attached.

For a SaaS business, swap Checkatrade for G2 and Capterra, add comparison and "alternatives" pages, and expect comparison sites and community threads to be cited ahead of you; the job is to be described accurately in them.

What not to do

Google's May 2026 guide warns that inauthentic mentions are a risk. Bought reviews, seeded Reddit threads, paid listicle placements dressed as editorial and networks of guest posts are detectable and they contaminate the footprint you are trying to build. The engines are looking for corroboration; fabricated corroboration is the one thing guaranteed to fail.

The plan, honestly stated

Fix access, rendering and the entity first, or the earned coverage will not connect to you. Then split effort roughly evenly between owned content that answers questions directly and earned presence in the five or six places your engines already cite for your questions. Ask the engines your buyer questions monthly and note which third-party domains they trust; that list is the target. Expect months, not weeks.

Your website is where the engine confirms what it has heard about you. It has to hear it somewhere first.